vm_machdep.c revision 182936
1/*- 2 * Copyright (c) 1982, 1986 The Regents of the University of California. 3 * Copyright (c) 1989, 1990 William Jolitz 4 * Copyright (c) 1994 John Dyson 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * the Systems Programming Group of the University of Utah Computer 9 * Science Department, and William Jolitz. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 41 */ 42 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: head/sys/amd64/amd64/vm_machdep.c 182936 2008-09-11 18:33:57Z jhb $"); 45 46#include "opt_isa.h" 47#include "opt_cpu.h" 48#include "opt_compat.h" 49 50#include <sys/param.h> 51#include <sys/systm.h> 52#include <sys/bio.h> 53#include <sys/buf.h> 54#include <sys/kernel.h> 55#include <sys/ktr.h> 56#include <sys/lock.h> 57#include <sys/malloc.h> 58#include <sys/mbuf.h> 59#include <sys/mutex.h> 60#include <sys/pioctl.h> 61#include <sys/proc.h> 62#include <sys/sf_buf.h> 63#include <sys/smp.h> 64#include <sys/sysctl.h> 65#include <sys/unistd.h> 66#include <sys/vnode.h> 67#include <sys/vmmeter.h> 68 69#include <machine/cpu.h> 70#include <machine/md_var.h> 71#include <machine/pcb.h> 72#include <machine/specialreg.h> 73 74#include <vm/vm.h> 75#include <vm/vm_extern.h> 76#include <vm/vm_kern.h> 77#include <vm/vm_page.h> 78#include <vm/vm_map.h> 79#include <vm/vm_param.h> 80 81#include <amd64/isa/isa.h> 82 83#ifdef COMPAT_IA32 84 85extern struct sysentvec ia32_freebsd_sysvec; 86 87#endif 88 89static void cpu_reset_real(void); 90#ifdef SMP 91static void cpu_reset_proxy(void); 92static u_int cpu_reset_proxyid; 93static volatile u_int cpu_reset_proxy_active; 94#endif 95 96/* 97 * Finish a fork operation, with process p2 nearly set up. 98 * Copy and update the pcb, set up the stack so that the child 99 * ready to run and return to user mode. 100 */ 101void 102cpu_fork(td1, p2, td2, flags) 103 register struct thread *td1; 104 register struct proc *p2; 105 struct thread *td2; 106 int flags; 107{ 108 register struct proc *p1; 109 struct pcb *pcb2; 110 struct mdproc *mdp2; 111 112 p1 = td1->td_proc; 113 if ((flags & RFPROC) == 0) 114 return; 115 116 /* Ensure that p1's pcb is up to date. */ 117 fpuexit(td1); 118 119 /* Point the pcb to the top of the stack */ 120 pcb2 = (struct pcb *)(td2->td_kstack + 121 td2->td_kstack_pages * PAGE_SIZE) - 1; 122 td2->td_pcb = pcb2; 123 124 /* Copy p1's pcb */ 125 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 126 127 /* Point mdproc and then copy over td1's contents */ 128 mdp2 = &p2->p_md; 129 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 130 131 /* 132 * Create a new fresh stack for the new process. 133 * Copy the trap frame for the return to user mode as if from a 134 * syscall. This copies most of the user mode register values. 135 */ 136 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 137 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 138 139 td2->td_frame->tf_rax = 0; /* Child returns zero */ 140 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 141 td2->td_frame->tf_rdx = 1; 142 143 /* 144 * If the parent process has the trap bit set (i.e. a debugger had 145 * single stepped the process to the system call), we need to clear 146 * the trap flag from the new frame unless the debugger had set PF_FORK 147 * on the parent. Otherwise, the child will receive a (likely 148 * unexpected) SIGTRAP when it executes the first instruction after 149 * returning to userland. 150 */ 151 if ((p1->p_pfsflags & PF_FORK) == 0) 152 td2->td_frame->tf_rflags &= ~PSL_T; 153 154 /* 155 * Set registers for trampoline to user mode. Leave space for the 156 * return address on stack. These are the kernel mode register values. 157 */ 158 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 159 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 160 pcb2->pcb_rbp = 0; 161 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 162 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 163 pcb2->pcb_rip = (register_t)fork_trampoline; 164 /*- 165 * pcb2->pcb_dr*: cloned above. 166 * pcb2->pcb_savefpu: cloned above. 167 * pcb2->pcb_flags: cloned above. 168 * pcb2->pcb_onfault: cloned above (always NULL here?). 169 * pcb2->pcb_[fg]sbase: cloned above 170 */ 171 172 /* Setup to release spin count in fork_exit(). */ 173 td2->td_md.md_spinlock_count = 1; 174 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 175 176 /* 177 * Now, cpu_switch() can schedule the new process. 178 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 179 * containing the return address when exiting cpu_switch. 180 * This will normally be to fork_trampoline(), which will have 181 * %ebx loaded with the new proc's pointer. fork_trampoline() 182 * will set up a stack to call fork_return(p, frame); to complete 183 * the return to user-mode. 184 */ 185} 186 187/* 188 * Intercept the return address from a freshly forked process that has NOT 189 * been scheduled yet. 190 * 191 * This is needed to make kernel threads stay in kernel mode. 192 */ 193void 194cpu_set_fork_handler(td, func, arg) 195 struct thread *td; 196 void (*func)(void *); 197 void *arg; 198{ 199 /* 200 * Note that the trap frame follows the args, so the function 201 * is really called like this: func(arg, frame); 202 */ 203 td->td_pcb->pcb_r12 = (long) func; /* function */ 204 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 205} 206 207void 208cpu_exit(struct thread *td) 209{ 210} 211 212void 213cpu_thread_exit(struct thread *td) 214{ 215 216 if (td == PCPU_GET(fpcurthread)) 217 fpudrop(); 218 219 /* Disable any hardware breakpoints. */ 220 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 221 reset_dbregs(); 222 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 223 } 224} 225 226void 227cpu_thread_clean(struct thread *td) 228{ 229} 230 231void 232cpu_thread_swapin(struct thread *td) 233{ 234} 235 236void 237cpu_thread_swapout(struct thread *td) 238{ 239} 240 241void 242cpu_thread_alloc(struct thread *td) 243{ 244 245 td->td_pcb = (struct pcb *)(td->td_kstack + 246 td->td_kstack_pages * PAGE_SIZE) - 1; 247 td->td_frame = (struct trapframe *)td->td_pcb - 1; 248} 249 250void 251cpu_thread_free(struct thread *td) 252{ 253} 254 255/* 256 * Initialize machine state (pcb and trap frame) for a new thread about to 257 * upcall. Put enough state in the new thread's PCB to get it to go back 258 * userret(), where we can intercept it again to set the return (upcall) 259 * Address and stack, along with those from upcals that are from other sources 260 * such as those generated in thread_userret() itself. 261 */ 262void 263cpu_set_upcall(struct thread *td, struct thread *td0) 264{ 265 struct pcb *pcb2; 266 267 /* Point the pcb to the top of the stack. */ 268 pcb2 = td->td_pcb; 269 270 /* 271 * Copy the upcall pcb. This loads kernel regs. 272 * Those not loaded individually below get their default 273 * values here. 274 */ 275 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 276 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 277 278 /* 279 * Create a new fresh stack for the new thread. 280 */ 281 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 282 283 /* 284 * Set registers for trampoline to user mode. Leave space for the 285 * return address on stack. These are the kernel mode register values. 286 */ 287 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 288 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 289 pcb2->pcb_rbp = 0; 290 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 291 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 292 pcb2->pcb_rip = (register_t)fork_trampoline; 293 /* 294 * If we didn't copy the pcb, we'd need to do the following registers: 295 * pcb2->pcb_dr*: cloned above. 296 * pcb2->pcb_savefpu: cloned above. 297 * pcb2->pcb_onfault: cloned above (always NULL here?). 298 * pcb2->pcb_[fg]sbase: cloned above 299 */ 300 301 /* Setup to release spin count in fork_exit(). */ 302 td->td_md.md_spinlock_count = 1; 303 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 304} 305 306/* 307 * Set that machine state for performing an upcall that has to 308 * be done in thread_userret() so that those upcalls generated 309 * in thread_userret() itself can be done as well. 310 */ 311void 312cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 313 stack_t *stack) 314{ 315 316 /* 317 * Do any extra cleaning that needs to be done. 318 * The thread may have optional components 319 * that are not present in a fresh thread. 320 * This may be a recycled thread so make it look 321 * as though it's newly allocated. 322 */ 323 cpu_thread_clean(td); 324 325#ifdef COMPAT_IA32 326 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) { 327 /* 328 * Set the trap frame to point at the beginning of the uts 329 * function. 330 */ 331 td->td_frame->tf_rbp = 0; 332 td->td_frame->tf_rsp = 333 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 334 td->td_frame->tf_rip = (uintptr_t)entry; 335 336 /* 337 * Pass the address of the mailbox for this kse to the uts 338 * function as a parameter on the stack. 339 */ 340 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 341 (uint32_t)(uintptr_t)arg); 342 343 return; 344 } 345#endif 346 347 /* 348 * Set the trap frame to point at the beginning of the uts 349 * function. 350 */ 351 td->td_frame->tf_rbp = 0; 352 td->td_frame->tf_rsp = 353 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 354 td->td_frame->tf_rsp -= 8; 355 td->td_frame->tf_rip = (register_t)entry; 356 357 /* 358 * Pass the address of the mailbox for this kse to the uts 359 * function as a parameter on the stack. 360 */ 361 td->td_frame->tf_rdi = (register_t)arg; 362} 363 364int 365cpu_set_user_tls(struct thread *td, void *tls_base) 366{ 367 368 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 369 return (EINVAL); 370 371#ifdef COMPAT_IA32 372 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) { 373 if (td == curthread) { 374 critical_enter(); 375 td->td_pcb->pcb_gsbase = (register_t)tls_base; 376 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase); 377 critical_exit(); 378 } else { 379 td->td_pcb->pcb_gsbase = (register_t)tls_base; 380 } 381 return (0); 382 } 383#endif 384 if (td == curthread) { 385 critical_enter(); 386 td->td_pcb->pcb_fsbase = (register_t)tls_base; 387 wrmsr(MSR_FSBASE, td->td_pcb->pcb_fsbase); 388 critical_exit(); 389 } else { 390 td->td_pcb->pcb_fsbase = (register_t)tls_base; 391 } 392 return (0); 393} 394 395#ifdef SMP 396static void 397cpu_reset_proxy() 398{ 399 400 cpu_reset_proxy_active = 1; 401 while (cpu_reset_proxy_active == 1) 402 ; /* Wait for other cpu to see that we've started */ 403 stop_cpus((1<<cpu_reset_proxyid)); 404 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 405 DELAY(1000000); 406 cpu_reset_real(); 407} 408#endif 409 410void 411cpu_reset() 412{ 413#ifdef SMP 414 u_int cnt, map; 415 416 if (smp_active) { 417 map = PCPU_GET(other_cpus) & ~stopped_cpus; 418 if (map != 0) { 419 printf("cpu_reset: Stopping other CPUs\n"); 420 stop_cpus(map); 421 } 422 423 if (PCPU_GET(cpuid) != 0) { 424 cpu_reset_proxyid = PCPU_GET(cpuid); 425 cpustop_restartfunc = cpu_reset_proxy; 426 cpu_reset_proxy_active = 0; 427 printf("cpu_reset: Restarting BSP\n"); 428 429 /* Restart CPU #0. */ 430 atomic_store_rel_int(&started_cpus, 1 << 0); 431 432 cnt = 0; 433 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 434 cnt++; /* Wait for BSP to announce restart */ 435 if (cpu_reset_proxy_active == 0) 436 printf("cpu_reset: Failed to restart BSP\n"); 437 enable_intr(); 438 cpu_reset_proxy_active = 2; 439 440 while (1); 441 /* NOTREACHED */ 442 } 443 444 DELAY(1000000); 445 } 446#endif 447 cpu_reset_real(); 448 /* NOTREACHED */ 449} 450 451static void 452cpu_reset_real() 453{ 454 struct region_descriptor null_idt; 455 int b; 456 457 disable_intr(); 458 459 /* 460 * Attempt to do a CPU reset via the keyboard controller, 461 * do not turn off GateA20, as any machine that fails 462 * to do the reset here would then end up in no man's land. 463 */ 464 outb(IO_KBD + 4, 0xFE); 465 DELAY(500000); /* wait 0.5 sec to see if that did it */ 466 467 /* 468 * Attempt to force a reset via the Reset Control register at 469 * I/O port 0xcf9. Bit 2 forces a system reset when it 470 * transitions from 0 to 1. Bit 1 selects the type of reset 471 * to attempt: 0 selects a "soft" reset, and 1 selects a 472 * "hard" reset. We try a "hard" reset. The first write sets 473 * bit 1 to select a "hard" reset and clears bit 2. The 474 * second write forces a 0 -> 1 transition in bit 2 to trigger 475 * a reset. 476 */ 477 outb(0xcf9, 0x2); 478 outb(0xcf9, 0x6); 479 DELAY(500000); /* wait 0.5 sec to see if that did it */ 480 481 /* 482 * Attempt to force a reset via the Fast A20 and Init register 483 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 484 * Bit 0 asserts INIT# when set to 1. We are careful to only 485 * preserve bit 1 while setting bit 0. We also must clear bit 486 * 0 before setting it if it isn't already clear. 487 */ 488 b = inb(0x92); 489 if (b != 0xff) { 490 if ((b & 0x1) != 0) 491 outb(0x92, b & 0xfe); 492 outb(0x92, b | 0x1); 493 DELAY(500000); /* wait 0.5 sec to see if that did it */ 494 } 495 496 printf("No known reset method worked, attempting CPU shutdown\n"); 497 DELAY(1000000); /* wait 1 sec for printf to complete */ 498 499 /* Wipe the IDT. */ 500 null_idt.rd_limit = 0; 501 null_idt.rd_base = 0; 502 lidt(&null_idt); 503 504 /* "good night, sweet prince .... <THUNK!>" */ 505 breakpoint(); 506 507 /* NOTREACHED */ 508 while(1); 509} 510 511/* 512 * Allocate an sf_buf for the given vm_page. On this machine, however, there 513 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 514 * returned. 515 */ 516struct sf_buf * 517sf_buf_alloc(struct vm_page *m, int pri) 518{ 519 520 return ((struct sf_buf *)m); 521} 522 523/* 524 * Free the sf_buf. In fact, do nothing because there are no resources 525 * associated with the sf_buf. 526 */ 527void 528sf_buf_free(struct sf_buf *sf) 529{ 530} 531 532/* 533 * Software interrupt handler for queued VM system processing. 534 */ 535void 536swi_vm(void *dummy) 537{ 538 if (busdma_swi_pending != 0) 539 busdma_swi(); 540} 541 542/* 543 * Tell whether this address is in some physical memory region. 544 * Currently used by the kernel coredump code in order to avoid 545 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 546 * or other unpredictable behaviour. 547 */ 548 549int 550is_physical_memory(vm_paddr_t addr) 551{ 552 553#ifdef DEV_ISA 554 /* The ISA ``memory hole''. */ 555 if (addr >= 0xa0000 && addr < 0x100000) 556 return 0; 557#endif 558 559 /* 560 * stuff other tests for known memory-mapped devices (PCI?) 561 * here 562 */ 563 564 return 1; 565} 566